Alexander Haneng's Post

Alexander Haneng · article · 2026-08-01 Reader 🖥️

You can replace (or supplement) a paid AI assistant like ChatGPT with a free, fully local, offline-capable setup — Google DeepMind's open-source Gemma 4 12B run through free LM Studio on ordinary consumer hardware like a MacBook M4 with 16GB RAM — at no real cost in quality, privacy, or convenience.

practical · mixed — A short Norwegian LinkedIn promo post, structured as a how-to 'recipe' (expository) in service of an implicit argumentative claim: run this local setup instead of / alongside paid cloud AI. There is no theoretical content — no explanation of how or why the model works, only a claimed outcome and a list of selling points.

Structure

Part	Role
Social-proof opener (281 people requested the guide)	credibility hook
Framing: this is a free, 100% private alternative to ChatGPT	thesis statement
Recipe: Gemma 4 12B (open-source, free) + LM Studio (free, cross-platform)	the how-to / method
Performance claim: on par with ChatGPT-4o-mini on a MacBook M4 (16GB)	evidence (asserted, not demonstrated)
Benefits list: free, private*, low power, offline-capable	supporting reasons / call to action
Asterisked caveat: data doesn't leave the PC 'unless you ask it to'	unexamined qualification undercutting the privacy claim

Key terms

oppskriften (the recipe)
A simple, replicable step-by-step formula for getting a ChatGPT-equivalent experience for free and privately.
100% privat
No data leaves the PC — but immediately qualified in the same post by an asterisk: 'unless you ask it to' (mechanism unspecified).
på høyde med ChatGPT 4o mini (on par with ChatGPT-4o-mini)
The author's own experience running Gemma 4 12B on a MacBook M4 (16GB) felt comparable in quality to ChatGPT-4o-mini, though no task, metric, or methodology is given.

Propositions

explicitcontestedRunning Gemma 4 12B via LM Studio on a MacBook M4 with 16GB RAM produces output on par with ChatGPT-4o-mini.conf: high
contradictsExperiences with local models for coding · p17 — This post treats an unbenchmarked, task-unspecified 'on par with ChatGPT-4o-mini' claim as meaningful and shareable, directly at odds with this corpus's own finding that such claims are close to meaningless without task and methodology.
contradictsLocal AI · p9 — Loukides's proposition that published benchmarks are an unreliable basis for model selection cuts directly against this post presenting an unsourced performance-parity claim as a settled fact.
supports95% cheaper and sovereign AI: Why we dropped Copilot at Coop Bank — LinkedIn post by Jannik William Frederiksen · p14 — Both posts assert that Gemma-family open-weight models are 'good enough' for real use — Frederiksen for enterprise knowledge-work at a bank, Haneng for a personal ChatGPT substitute — though neither provides benchmark evidence.
explicitRunning an AI language model locally is much easier than most people think, and has a range of advantages.conf: high
explicitThis local setup is free, keeps data on the device (unless the model is explicitly asked to send it out), requires no internet connection, and draws low power (only the local machine, no data center).conf: high
implicitOrdinary consumer hardware (a MacBook M4 with 16GB RAM) is now sufficient to match commercial cloud-model quality for the reader's purposes, without qualification by task type.conf: medium
supportsGemma 4 QAT models: Optimizing model compression for mobile and laptop efficiency · p8 — Both documents assert, from different angles, that on-device/local LLM inference on consumer hardware has become a practical default rather than a niche or research pursuit.
implicitA capability claim asserted without benchmark data, task specification, or methodology is credible enough to act on and share as a 'recipe' others should follow.conf: medium

Arguments

You should run Gemma 4 12B locally via LM Studio on your own machine instead of, or alongside, a paid cloud AI assistant like ChatGPT. strength: low

Explicit premises:

Both LM Studio and Gemma 4 12B are free and open-source.
The setup is private — no data leaves the PC (with a footnoted exception).
Performance on a MacBook M4 (16GB) is on par with ChatGPT-4o-mini.
It works without an internet connection.
Power consumption is low (only the local machine).

Implicit premises:

The reader has access to hardware roughly equivalent to a MacBook M4 with 16GB RAM.
'On par with ChatGPT-4o-mini' is a meaningful, task-independent measure that generalizes to the reader's own use cases.
An unbenchmarked, single-anecdote capability claim from a promotional LinkedIn post is a sufficient basis for an adoption decision.
The privacy guarantee holds by default, and the mechanism by which data could otherwise leave (the asterisked exception) doesn't need explaining for the claim to count as '100% private.'

Inferences:

If performance really is on par with 4o-mini and the setup is free, private, and offline-capable, then a paid cloud AI subscription becomes largely redundant for many everyday uses — an implication the post gestures at (with the 'vibe coding on the plane' aside) but never states outright.

Narratives

281 people reportedly asked to have this guide sent to them (opening social-proof line).
→ Substitutes popularity/virality for evidence, implicitly supporting the post's credibility (the implicit premise that an unbenchmarked claim is trustworthy) without adding any data about the claim itself.

Questions addressed

✅ How can someone get a ChatGPT-like AI experience for free and with full privacy?
Run Gemma 4 12B through LM Studio locally on consumer hardware such as a MacBook M4 with 16GB RAM.
⭕ Does the claimed performance parity with ChatGPT-4o-mini actually hold — across which tasks, measured how?

Gaps

incompleteThe post's central claim — Gemma 4 12B on a MacBook M4 (16GB) via LM Studio is 'på høyde med' (on par with) ChatGPT-4o-mini — carries no benchmark, no named task or use case, and no methodology. This is exactly the pattern this corpus has already flagged elsewhere as epistemically empty: benchmark-style claims without task specification and methodology are close to meaningless for small/local models, and published benchmarks are an unreliable basis for model selection without domain-specific evaluation.
Read next: extract-experiences-with-local-models-for-coding (a rigorous, task-by-task account of the same class of claim) and extract-local-ai-loukides (explicit critique of benchmark-based model selection).
incomplete'100% privat' is asserted, then immediately undercut by a footnote ('*Om du ikke ber den om det da' — unless you ask it to) with no explanation of the mechanism: does LM Studio's default configuration ever give Gemma network or tool access? Is the caveat about explicit web-search plugins, MCP-style tool use, or something else? Without that, the reader can't assess what the actual privacy guarantee is.
Read next: None in this corpus directly explains LM Studio's tool/network-access model; would need vendor documentation.
incompleteThe post treats a single, unspecified use case (implicitly 'chatting') as representative, when this corpus's own hands-on local-model accounts show viability is highly task- and hardware-dependent — the same model and settings can produce meaningfully different quality on different machines, and small local models handle simple tasks acceptably but fall apart on complex ones.
Read next: extract-viability-of-local-models-for-coding and extract-experiences-with-local-models-for-coding.